This paper describes an experimental set-up for the predictive diagnosis of faults in mechanical systems and in particular in gearing and shows the main results obtained. The basic idea is to set up a diagnostic tool capable of recognizing faults in power trains such as damages on gear teeth. Moreover the test rig was designed to allow to vary the relative distance and angle between the two gears shafts. The procedure starts by recording “the signature” of the healthy gearing (i.e. still-undamaged and integer). The systematic analysis of the signals and the subsequent comparison with the features corresponding to the failure conditions is then taken into account. Two different vibration-based monitoring techniques are taken into account; spectrum and cepstrum analyses are compared, using the same vibration data measurements under different operating conditions, on the same gearbox. The signals are obtained through, encoders, accelerometers and optical distance sensors.